America’s economic rebound despite tariffs during the second quarter is not a mystery — it was all dependent on artificial intelligence and data center construction.
Real GDP grew at a brisk 3.8% annualized in Q2 2025 and is expected to grow at least 2.5% in Q3 (almost 4% again according to Atlanta’s GDP Now), but the blunt truth is that the U.S. expansion increasingly rides on high-end consumers spending and more importantly on artificial intelligence and the physical plants that power it: data centers, substations, high-voltage lines, and the skilled trades who make these massive assets perform.
My thesis is simple: if the United States wants to sustain this growth over the next 12 months and avoid a natural slowdown, it must go all-in now on building AI infrastructure — slash permitting timelines, green-light energy projects, and mass-train the HVAC techs, electricians, and plumbers who turn capital into capacity. The Office of Science and Technology Policy at the White House recently issued a new Request for Information to gather ideas from experts and think tanks to deregulate the sector.
Consider the scale. Primary U.S. data-center markets had roughly 6.35 GW under construction at end-2024, more than double the year before — a pipeline translating to about $74 billion in construction investment excluding land and software. CBRE corroborates the doubling: North American supply under construction hit a record 6,350 MW in 2024, a twelve-fold leap from 2020. Communications construction outlays are now averaging $2.4 billion per month in the U.S., up 25% versus the two years before ChatGPT’s release; and analysts do not foresee a pullback. Vacancies in the tightest markets — Northern Virginia sub-1% — underline the mismatch between surging demand and the grid’s ability to deliver power where compute wants to cluster.
But did AI move the macro needle in Q2? Yes — through investment. Bureau of Economic Analysis BEA’s third estimate confirms the 3.8% pace; while the quarter’s composition reflected a swing in trade and strong consumption, multiple market analyses point to AI-related capex as a distinct tailwind in the first half of 2025. JPMorgan estimates that AI spending contributed about 1.1 percentage points to first-half GDP growth — the equivalent of the U.S. consumer. The AI build-out shows up across equipment, IP, and communication structures — precisely the categories now scaling fastest.
The broader energy system is racing to keep up. After a decade of flat demand, U.S. electricity use is setting records, with the EIA projecting 4,191 TWh in 2025 and 4,305 TWh in 2026, citing data centers as a key driver. The EIA’s Annual Energy Outlook highlights computing as the fastest-rising end use in commercial buildings; by 2050, computing alone could account for one-fifth of commercial electricity consumption. Deloitte’s utility analysis is blunter: AI-center power demand could climb from ~4 GW in 2024 to 123 GW by 2035. In short, the constraint to AI is no longer imagination or capital; it is permitting, power, and people.
On permitting, we are stuck in a 20th-century gridlock. The result is a stealth moratorium: multi-year waits for interconnection, environmental reviews that sprawl beyond their mandate, and local zoning that arbitrages away national competitiveness. The policy remedy is purely administrative here.
First, federal and state categorical exclusions should cover routine grid upgrades and standard-design data-center shells with pre-certified efficiency and noise profiles. Second, shot-clock deadlines must apply to interconnection studies and right-of-way approvals, with automatic approvals if agencies blow the clock. The White House has already signaled an interest in expedition; it should be codified and widened, with clear guardrails on siting and emissions intensity.
On power, the lesson from Europe’s moratoria is cautionary: where capacity is scarce and approvals slow, investment doesn’t pause — it just re-routes to other countries. The U.S. should take the opposite track. Prioritize grid-serving transmission in and out of the top AI hubs; accelerate transformer manufacturing; and pair new campuses with dedicated clean power (utility-scale solar, wind, nuclear uprates, storage) under streamlined contracts. AI is often portrayed as an emissions problem; it can instead be the anchor tenant for the clean-energy build we have long deferred.
On people, the constraint is acute. Construction firms report a shortfall of 0.5–0.75 million workers, with skilled trades in particularly short supply; wages are outrunning the broader economy as firms train on the job. Data centers intensify this scarcity: high-spec HVAC, power distribution, specialty welding, commissioning — every task requires specific talents. A rational response is a wartime-style training surge: fast-track credentialing for electricians and HVAC specialists; paid apprenticeships under project labor agreements; federal matching grants for community-college programs tied to hyperscale campuses. If we can subsidize semiconductors, we can co-finance the human capital that makes chips useful.
RESTORING AMERICA: HARNESSING AI TO MAKE AMERICA HEALTHY AGAIN
We have been here before. America’s 19th-century railroad boom and 20th-century highway system were growth engines that reorganized production and settlement. 30 years ago, we built the massive infrastructure underpinning the new internet. Today’s rails are fiber and power lines; today’s junctions are substation yards. Generative AI has become a structural driver of construction, with U.S. communications outlays up a quarter since the pre-ChatGPT baseline and tens of billions now in the data-center pipeline — momentum that will persist amid continued AI enthusiasm unless choked by grid and permitting bottlenecks.
If we want an economy that compounds, we must build the compute that compounds — and the wires, transformers, switchgear, cooling loops, and human skill to match. Relax the chokepoints and free all projects.
Sebastien Laye is an economist and AI entrepreneur.
